有一个主表left join 同一个小表两次分页语句，因为order by 导致执行时做排序，从执行计划中Using filesort ，以及profile中creating sort index 耗时可以看出。

从trace文件可以看出filesort的计算：

"join_execution": {
        "select#": 1,
        "steps": [
          {
            "filesort_information": [
              {
                "direction": "desc",
                "table": "`topxxx` `t`",
                "field": "create_date"
              }
            ] /* filesort_information */,
            "filesort_priority_queue_optimization": {
              "limit": 20,
              "rows_estimate": 2302749,
              "row_size": 264,
              "memory_available": 4194304,
              "chosen": true
            } /* filesort_priority_queue_optimization */,
            "filesort_execution": [
            ] /* filesort_execution */,
            "filesort_summary": {
              "rows": 21,
              "examined_rows": 216594,
              "number_of_tmp_files": 0,
              "sort_buffer_size": 5712,
              "sort_mode": ""
            } /* filesort_summary */
          }
        ] /* steps */
      } /* join_execution */
    }

后面通过索引加入排序字段后减去排序操作,排序字段放在索引的最前面。

create index idx_topxxx1 on topic (create_date desc,is_del,is_en);

trace 中可以看出排序使用了索引。

            "reconsidering_access_paths_for_index_ordering": {
              "clause": "ORDER BY",
              "index_order_summary": {
                "table": "`topic` `t`",
                "index_provides_order": true,
                "order_direction": "desc",
                "index": "idx_topxxx1",
                "plan_changed": true,
                "access_type": "index"
              } /* index_order_summary */
            } /* reconsidering_access_paths_for_index_ordering */
          },
          {

以此记录。